git.oblomov.eu Git - linux-2.6/blob - net/core/sock.c

   1 /*
   2  * INET         An implementation of the TCP/IP protocol suite for the LINUX
   3  *              operating system.  INET is implemented using the  BSD Socket
   4  *              interface as the means of communication with the user level.
   5  *
   6  *              Generic socket support routines. Memory allocators, socket lock/release
   7  *              handler for protocols to use and generic option handler.
   8  *
   9  *
  10  * Version:     $Id: sock.c,v 1.117 2002/02/01 22:01:03 davem Exp $
  11  *
  12  * Authors:     Ross Biro
  13  *              Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
  14  *              Florian La Roche, <flla@stud.uni-sb.de>
  15  *              Alan Cox, <A.Cox@swansea.ac.uk>
  16  *
  17  * Fixes:
  18  *              Alan Cox        :       Numerous verify_area() problems
  19  *              Alan Cox        :       Connecting on a connecting socket
  20  *                                      now returns an error for tcp.
  21  *              Alan Cox        :       sock->protocol is set correctly.
  22  *                                      and is not sometimes left as 0.
  23  *              Alan Cox        :       connect handles icmp errors on a
  24  *                                      connect properly. Unfortunately there
  25  *                                      is a restart syscall nasty there. I
  26  *                                      can't match BSD without hacking the C
  27  *                                      library. Ideas urgently sought!
  28  *              Alan Cox        :       Disallow bind() to addresses that are
  29  *                                      not ours - especially broadcast ones!!
  30  *              Alan Cox        :       Socket 1024 _IS_ ok for users. (fencepost)
  31  *              Alan Cox        :       sock_wfree/sock_rfree don't destroy sockets,
  32  *                                      instead they leave that for the DESTROY timer.
  33  *              Alan Cox        :       Clean up error flag in accept
  34  *              Alan Cox        :       TCP ack handling is buggy, the DESTROY timer
  35  *                                      was buggy. Put a remove_sock() in the handler
  36  *                                      for memory when we hit 0. Also altered the timer
  37  *                                      code. The ACK stuff can wait and needs major
  38  *                                      TCP layer surgery.
  39  *              Alan Cox        :       Fixed TCP ack bug, removed remove sock
  40  *                                      and fixed timer/inet_bh race.
  41  *              Alan Cox        :       Added zapped flag for TCP
  42  *              Alan Cox        :       Move kfree_skb into skbuff.c and tidied up surplus code
  43  *              Alan Cox        :       for new sk_buff allocations wmalloc/rmalloc now call alloc_skb
  44  *              Alan Cox        :       kfree_s calls now are kfree_skbmem so we can track skb resources
  45  *              Alan Cox        :       Supports socket option broadcast now as does udp. Packet and raw need fixing.
  46  *              Alan Cox        :       Added RCVBUF,SNDBUF size setting. It suddenly occurred to me how easy it was so...
  47  *              Rick Sladkey    :       Relaxed UDP rules for matching packets.
  48  *              C.E.Hawkins     :       IFF_PROMISC/SIOCGHWADDR support
  49  *      Pauline Middelink       :       identd support
  50  *              Alan Cox        :       Fixed connect() taking signals I think.
  51  *              Alan Cox        :       SO_LINGER supported
  52  *              Alan Cox        :       Error reporting fixes
  53  *              Anonymous       :       inet_create tidied up (sk->reuse setting)
  54  *              Alan Cox        :       inet sockets don't set sk->type!
  55  *              Alan Cox        :       Split socket option code
  56  *              Alan Cox        :       Callbacks
  57  *              Alan Cox        :       Nagle flag for Charles & Johannes stuff
  58  *              Alex            :       Removed restriction on inet fioctl
  59  *              Alan Cox        :       Splitting INET from NET core
  60  *              Alan Cox        :       Fixed bogus SO_TYPE handling in getsockopt()
  61  *              Adam Caldwell   :       Missing return in SO_DONTROUTE/SO_DEBUG code
  62  *              Alan Cox        :       Split IP from generic code
  63  *              Alan Cox        :       New kfree_skbmem()
  64  *              Alan Cox        :       Make SO_DEBUG superuser only.
  65  *              Alan Cox        :       Allow anyone to clear SO_DEBUG
  66  *                                      (compatibility fix)
  67  *              Alan Cox        :       Added optimistic memory grabbing for AF_UNIX throughput.
  68  *              Alan Cox        :       Allocator for a socket is settable.
  69  *              Alan Cox        :       SO_ERROR includes soft errors.
  70  *              Alan Cox        :       Allow NULL arguments on some SO_ opts
  71  *              Alan Cox        :       Generic socket allocation to make hooks
  72  *                                      easier (suggested by Craig Metz).
  73  *              Michael Pall    :       SO_ERROR returns positive errno again
  74  *              Steve Whitehouse:       Added default destructor to free
  75  *                                      protocol private data.
  76  *              Steve Whitehouse:       Added various other default routines
  77  *                                      common to several socket families.
  78  *              Chris Evans     :       Call suser() check last on F_SETOWN
  79  *              Jay Schulist    :       Added SO_ATTACH_FILTER and SO_DETACH_FILTER.
  80  *              Andi Kleen      :       Add sock_kmalloc()/sock_kfree_s()
  81  *              Andi Kleen      :       Fix write_space callback
  82  *              Chris Evans     :       Security fixes - signedness again
  83  *              Arnaldo C. Melo :       cleanups, use skb_queue_purge
  84  *
  85  * To Fix:
  86  *
  87  *
  88  *              This program is free software; you can redistribute it and/or
  89  *              modify it under the terms of the GNU General Public License
  90  *              as published by the Free Software Foundation; either version
  91  *              2 of the License, or (at your option) any later version.
  92  */
  93
  94 #include <linux/capability.h>
  95 #include <linux/config.h>
  96 #include <linux/errno.h>
  97 #include <linux/types.h>
  98 #include <linux/socket.h>
  99 #include <linux/in.h>
 100 #include <linux/kernel.h>
 101 #include <linux/module.h>
 102 #include <linux/proc_fs.h>
 103 #include <linux/seq_file.h>
 104 #include <linux/sched.h>
 105 #include <linux/timer.h>
 106 #include <linux/string.h>
 107 #include <linux/sockios.h>
 108 #include <linux/net.h>
 109 #include <linux/mm.h>
 110 #include <linux/slab.h>
 111 #include <linux/interrupt.h>
 112 #include <linux/poll.h>
 113 #include <linux/tcp.h>
 114 #include <linux/init.h>
 115
 116 #include <asm/uaccess.h>
 117 #include <asm/system.h>
 118
 119 #include <linux/netdevice.h>
 120 #include <net/protocol.h>
 121 #include <linux/skbuff.h>
 122 #include <net/request_sock.h>
 123 #include <net/sock.h>
 124 #include <net/xfrm.h>
 125 #include <linux/ipsec.h>
 126
 127 #include <linux/filter.h>
 128
 129 #ifdef CONFIG_INET
 130 #include <net/tcp.h>
 131 #endif
 132
 133 /* Take into consideration the size of the struct sk_buff overhead in the
 134  * determination of these values, since that is non-constant across
 135  * platforms.  This makes socket queueing behavior and performance
 136  * not depend upon such differences.
 137  */
 138 #define _SK_MEM_PACKETS         256
 139 #define _SK_MEM_OVERHEAD        (sizeof(struct sk_buff) + 256)
 140 #define SK_WMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 141 #define SK_RMEM_MAX             (_SK_MEM_OVERHEAD * _SK_MEM_PACKETS)
 142
 143 /* Run time adjustable parameters. */
 144 __u32 sysctl_wmem_max = SK_WMEM_MAX;
 145 __u32 sysctl_rmem_max = SK_RMEM_MAX;
 146 __u32 sysctl_wmem_default = SK_WMEM_MAX;
 147 __u32 sysctl_rmem_default = SK_RMEM_MAX;
 148
 149 /* Maximal space eaten by iovec or ancilliary data plus some space */
 150 int sysctl_optmem_max = sizeof(unsigned long)*(2*UIO_MAXIOV + 512);
 151
 152 static int sock_set_timeout(long *timeo_p, char __user *optval, int optlen)
 153 {
 154         struct timeval tv;
 155
 156         if (optlen < sizeof(tv))
 157                 return -EINVAL;
 158         if (copy_from_user(&tv, optval, sizeof(tv)))
 159                 return -EFAULT;
 160
 161         *timeo_p = MAX_SCHEDULE_TIMEOUT;
 162         if (tv.tv_sec == 0 && tv.tv_usec == 0)
 163                 return 0;
 164         if (tv.tv_sec < (MAX_SCHEDULE_TIMEOUT/HZ - 1))
 165                 *timeo_p = tv.tv_sec*HZ + (tv.tv_usec+(1000000/HZ-1))/(1000000/HZ);
 166         return 0;
 167 }
 168
 169 static void sock_warn_obsolete_bsdism(const char *name)
 170 {
 171         static int warned;
 172         static char warncomm[TASK_COMM_LEN];
 173         if (strcmp(warncomm, current->comm) && warned < 5) {
 174                 strcpy(warncomm,  current->comm);
 175                 printk(KERN_WARNING "process `%s' is using obsolete "
 176                        "%s SO_BSDCOMPAT\n", warncomm, name);
 177                 warned++;
 178         }
 179 }
 180
 181 static void sock_disable_timestamp(struct sock *sk)
 182 {
 183         if (sock_flag(sk, SOCK_TIMESTAMP)) {
 184                 sock_reset_flag(sk, SOCK_TIMESTAMP);
 185                 net_disable_timestamp();
 186         }
 187 }
 188
 189
 190 /*
 191  *      This is meant for all protocols to use and covers goings on
 192  *      at the socket level. Everything here is generic.
 193  */
 194
 195 int sock_setsockopt(struct socket *sock, int level, int optname,
 196                     char __user *optval, int optlen)
 197 {
 198         struct sock *sk=sock->sk;
 199         struct sk_filter *filter;
 200         int val;
 201         int valbool;
 202         struct linger ling;
 203         int ret = 0;
 204
 205         /*
 206          *      Options without arguments
 207          */
 208
 209 #ifdef SO_DONTLINGER            /* Compatibility item... */
 210         if (optname == SO_DONTLINGER) {
 211                 lock_sock(sk);
 212                 sock_reset_flag(sk, SOCK_LINGER);
 213                 release_sock(sk);
 214                 return 0;
 215         }
 216 #endif
 217
 218         if(optlen<sizeof(int))
 219                 return(-EINVAL);
 220
 221         if (get_user(val, (int __user *)optval))
 222                 return -EFAULT;
 223
 224         valbool = val?1:0;
 225
 226         lock_sock(sk);
 227
 228         switch(optname)
 229         {
 230                 case SO_DEBUG:
 231                         if(val && !capable(CAP_NET_ADMIN))
 232                         {
 233                                 ret = -EACCES;
 234                         }
 235                         else if (valbool)
 236                                 sock_set_flag(sk, SOCK_DBG);
 237                         else
 238                                 sock_reset_flag(sk, SOCK_DBG);
 239                         break;
 240                 case SO_REUSEADDR:
 241                         sk->sk_reuse = valbool;
 242                         break;
 243                 case SO_TYPE:
 244                 case SO_ERROR:
 245                         ret = -ENOPROTOOPT;
 246                         break;
 247                 case SO_DONTROUTE:
 248                         if (valbool)
 249                                 sock_set_flag(sk, SOCK_LOCALROUTE);
 250                         else
 251                                 sock_reset_flag(sk, SOCK_LOCALROUTE);
 252                         break;
 253                 case SO_BROADCAST:
 254                         sock_valbool_flag(sk, SOCK_BROADCAST, valbool);
 255                         break;
 256                 case SO_SNDBUF:
 257                         /* Don't error on this BSD doesn't and if you think
 258                            about it this is right. Otherwise apps have to
 259                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 260                            are treated in BSD as hints */
 261
 262                         if (val > sysctl_wmem_max)
 263                                 val = sysctl_wmem_max;
 264 set_sndbuf:
 265                         sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
 266                         if ((val * 2) < SOCK_MIN_SNDBUF)
 267                                 sk->sk_sndbuf = SOCK_MIN_SNDBUF;
 268                         else
 269                                 sk->sk_sndbuf = val * 2;
 270
 271                         /*
 272                          *      Wake up sending tasks if we
 273                          *      upped the value.
 274                          */
 275                         sk->sk_write_space(sk);
 276                         break;
 277
 278                 case SO_SNDBUFFORCE:
 279                         if (!capable(CAP_NET_ADMIN)) {
 280                                 ret = -EPERM;
 281                                 break;
 282                         }
 283                         goto set_sndbuf;
 284
 285                 case SO_RCVBUF:
 286                         /* Don't error on this BSD doesn't and if you think
 287                            about it this is right. Otherwise apps have to
 288                            play 'guess the biggest size' games. RCVBUF/SNDBUF
 289                            are treated in BSD as hints */
 290
 291                         if (val > sysctl_rmem_max)
 292                                 val = sysctl_rmem_max;
 293 set_rcvbuf:
 294                         sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
 295                         /* FIXME: is this lower bound the right one? */
 296                         if ((val * 2) < SOCK_MIN_RCVBUF)
 297                                 sk->sk_rcvbuf = SOCK_MIN_RCVBUF;
 298                         else
 299                                 sk->sk_rcvbuf = val * 2;
 300                         break;
 301
 302                 case SO_RCVBUFFORCE:
 303                         if (!capable(CAP_NET_ADMIN)) {
 304                                 ret = -EPERM;
 305                                 break;
 306                         }
 307                         goto set_rcvbuf;
 308
 309                 case SO_KEEPALIVE:
 310 #ifdef CONFIG_INET
 311                         if (sk->sk_protocol == IPPROTO_TCP)
 312                                 tcp_set_keepalive(sk, valbool);
 313 #endif
 314                         sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool);
 315                         break;
 316
 317                 case SO_OOBINLINE:
 318                         sock_valbool_flag(sk, SOCK_URGINLINE, valbool);
 319                         break;
 320
 321                 case SO_NO_CHECK:
 322                         sk->sk_no_check = valbool;
 323                         break;
 324
 325                 case SO_PRIORITY:
 326                         if ((val >= 0 && val <= 6) || capable(CAP_NET_ADMIN))
 327                                 sk->sk_priority = val;
 328                         else
 329                                 ret = -EPERM;
 330                         break;
 331
 332                 case SO_LINGER:
 333                         if(optlen<sizeof(ling)) {
 334                                 ret = -EINVAL;  /* 1003.1g */
 335                                 break;
 336                         }
 337                         if (copy_from_user(&ling,optval,sizeof(ling))) {
 338                                 ret = -EFAULT;
 339                                 break;
 340                         }
 341                         if (!ling.l_onoff)
 342                                 sock_reset_flag(sk, SOCK_LINGER);
 343                         else {
 344 #if (BITS_PER_LONG == 32)
 345                                 if ((unsigned int)ling.l_linger >= MAX_SCHEDULE_TIMEOUT/HZ)
 346                                         sk->sk_lingertime = MAX_SCHEDULE_TIMEOUT;
 347                                 else
 348 #endif
 349                                         sk->sk_lingertime = (unsigned int)ling.l_linger * HZ;
 350                                 sock_set_flag(sk, SOCK_LINGER);
 351                         }
 352                         break;
 353
 354                 case SO_BSDCOMPAT:
 355                         sock_warn_obsolete_bsdism("setsockopt");
 356                         break;
 357
 358                 case SO_PASSCRED:
 359                         if (valbool)
 360                                 set_bit(SOCK_PASSCRED, &sock->flags);
 361                         else
 362                                 clear_bit(SOCK_PASSCRED, &sock->flags);
 363                         break;
 364
 365                 case SO_TIMESTAMP:
 366                         if (valbool)  {
 367                                 sock_set_flag(sk, SOCK_RCVTSTAMP);
 368                                 sock_enable_timestamp(sk);
 369                         } else
 370                                 sock_reset_flag(sk, SOCK_RCVTSTAMP);
 371                         break;
 372
 373                 case SO_RCVLOWAT:
 374                         if (val < 0)
 375                                 val = INT_MAX;
 376                         sk->sk_rcvlowat = val ? : 1;
 377                         break;
 378
 379                 case SO_RCVTIMEO:
 380                         ret = sock_set_timeout(&sk->sk_rcvtimeo, optval, optlen);
 381                         break;
 382
 383                 case SO_SNDTIMEO:
 384                         ret = sock_set_timeout(&sk->sk_sndtimeo, optval, optlen);
 385                         break;
 386
 387 #ifdef CONFIG_NETDEVICES
 388                 case SO_BINDTODEVICE:
 389                 {
 390                         char devname[IFNAMSIZ];
 391
 392                         /* Sorry... */
 393                         if (!capable(CAP_NET_RAW)) {
 394                                 ret = -EPERM;
 395                                 break;
 396                         }
 397
 398                         /* Bind this socket to a particular device like "eth0",
 399                          * as specified in the passed interface name. If the
 400                          * name is "" or the option length is zero the socket
 401                          * is not bound.
 402                          */
 403
 404                         if (!valbool) {
 405                                 sk->sk_bound_dev_if = 0;
 406                         } else {
 407                                 if (optlen > IFNAMSIZ)
 408                                         optlen = IFNAMSIZ;
 409                                 if (copy_from_user(devname, optval, optlen)) {
 410                                         ret = -EFAULT;
 411                                         break;
 412                                 }
 413
 414                                 /* Remove any cached route for this socket. */
 415                                 sk_dst_reset(sk);
 416
 417                                 if (devname[0] == '\0') {
 418                                         sk->sk_bound_dev_if = 0;
 419                                 } else {
 420                                         struct net_device *dev = dev_get_by_name(devname);
 421                                         if (!dev) {
 422                                                 ret = -ENODEV;
 423                                                 break;
 424                                         }
 425                                         sk->sk_bound_dev_if = dev->ifindex;
 426                                         dev_put(dev);
 427                                 }
 428                         }
 429                         break;
 430                 }
 431 #endif
 432
 433
 434                 case SO_ATTACH_FILTER:
 435                         ret = -EINVAL;
 436                         if (optlen == sizeof(struct sock_fprog)) {
 437                                 struct sock_fprog fprog;
 438
 439                                 ret = -EFAULT;
 440                                 if (copy_from_user(&fprog, optval, sizeof(fprog)))
 441                                         break;
 442
 443                                 ret = sk_attach_filter(&fprog, sk);
 444                         }
 445                         break;
 446
 447                 case SO_DETACH_FILTER:
 448                         spin_lock_bh(&sk->sk_lock.slock);
 449                         filter = sk->sk_filter;
 450                         if (filter) {
 451                                 sk->sk_filter = NULL;
 452                                 spin_unlock_bh(&sk->sk_lock.slock);
 453                                 sk_filter_release(sk, filter);
 454                                 break;
 455                         }
 456                         spin_unlock_bh(&sk->sk_lock.slock);
 457                         ret = -ENONET;
 458                         break;
 459
 460                 /* We implement the SO_SNDLOWAT etc to
 461                    not be settable (1003.1g 5.3) */
 462                 default:
 463                         ret = -ENOPROTOOPT;
 464                         break;
 465         }
 466         release_sock(sk);
 467         return ret;
 468 }
 469
 470
 471 int sock_getsockopt(struct socket *sock, int level, int optname,
 472                     char __user *optval, int __user *optlen)
 473 {
 474         struct sock *sk = sock->sk;
 475
 476         union
 477         {
 478                 int val;
 479                 struct linger ling;
 480                 struct timeval tm;
 481         } v;
 482
 483         unsigned int lv = sizeof(int);
 484         int len;
 485
 486         if(get_user(len,optlen))
 487                 return -EFAULT;
 488         if(len < 0)
 489                 return -EINVAL;
 490
 491         switch(optname)
 492         {
 493                 case SO_DEBUG:
 494                         v.val = sock_flag(sk, SOCK_DBG);
 495                         break;
 496
 497                 case SO_DONTROUTE:
 498                         v.val = sock_flag(sk, SOCK_LOCALROUTE);
 499                         break;
 500
 501                 case SO_BROADCAST:
 502                         v.val = !!sock_flag(sk, SOCK_BROADCAST);
 503                         break;
 504
 505                 case SO_SNDBUF:
 506                         v.val = sk->sk_sndbuf;
 507                         break;
 508
 509                 case SO_RCVBUF:
 510                         v.val = sk->sk_rcvbuf;
 511                         break;
 512
 513                 case SO_REUSEADDR:
 514                         v.val = sk->sk_reuse;
 515                         break;
 516
 517                 case SO_KEEPALIVE:
 518                         v.val = !!sock_flag(sk, SOCK_KEEPOPEN);
 519                         break;
 520
 521                 case SO_TYPE:
 522                         v.val = sk->sk_type;
 523                         break;
 524
 525                 case SO_ERROR:
 526                         v.val = -sock_error(sk);
 527                         if(v.val==0)
 528                                 v.val = xchg(&sk->sk_err_soft, 0);
 529                         break;
 530
 531                 case SO_OOBINLINE:
 532                         v.val = !!sock_flag(sk, SOCK_URGINLINE);
 533                         break;
 534
 535                 case SO_NO_CHECK:
 536                         v.val = sk->sk_no_check;
 537                         break;
 538
 539                 case SO_PRIORITY:
 540                         v.val = sk->sk_priority;
 541                         break;
 542
 543                 case SO_LINGER:
 544                         lv              = sizeof(v.ling);
 545                         v.ling.l_onoff  = !!sock_flag(sk, SOCK_LINGER);
 546                         v.ling.l_linger = sk->sk_lingertime / HZ;
 547                         break;
 548
 549                 case SO_BSDCOMPAT:
 550                         sock_warn_obsolete_bsdism("getsockopt");
 551                         break;
 552
 553                 case SO_TIMESTAMP:
 554                         v.val = sock_flag(sk, SOCK_RCVTSTAMP);
 555                         break;
 556
 557                 case SO_RCVTIMEO:
 558                         lv=sizeof(struct timeval);
 559                         if (sk->sk_rcvtimeo == MAX_SCHEDULE_TIMEOUT) {
 560                                 v.tm.tv_sec = 0;
 561                                 v.tm.tv_usec = 0;
 562                         } else {
 563                                 v.tm.tv_sec = sk->sk_rcvtimeo / HZ;
 564                                 v.tm.tv_usec = ((sk->sk_rcvtimeo % HZ) * 1000000) / HZ;
 565                         }
 566                         break;
 567
 568                 case SO_SNDTIMEO:
 569                         lv=sizeof(struct timeval);
 570                         if (sk->sk_sndtimeo == MAX_SCHEDULE_TIMEOUT) {
 571                                 v.tm.tv_sec = 0;
 572                                 v.tm.tv_usec = 0;
 573                         } else {
 574                                 v.tm.tv_sec = sk->sk_sndtimeo / HZ;
 575                                 v.tm.tv_usec = ((sk->sk_sndtimeo % HZ) * 1000000) / HZ;
 576                         }
 577                         break;
 578
 579                 case SO_RCVLOWAT:
 580                         v.val = sk->sk_rcvlowat;
 581                         break;
 582
 583                 case SO_SNDLOWAT:
 584                         v.val=1;
 585                         break;
 586
 587                 case SO_PASSCRED:
 588                         v.val = test_bit(SOCK_PASSCRED, &sock->flags) ? 1 : 0;
 589                         break;
 590
 591                 case SO_PEERCRED:
 592                         if (len > sizeof(sk->sk_peercred))
 593                                 len = sizeof(sk->sk_peercred);
 594                         if (copy_to_user(optval, &sk->sk_peercred, len))
 595                                 return -EFAULT;
 596                         goto lenout;
 597
 598                 case SO_PEERNAME:
 599                 {
 600                         char address[128];
 601
 602                         if (sock->ops->getname(sock, (struct sockaddr *)address, &lv, 2))
 603                                 return -ENOTCONN;
 604                         if (lv < len)
 605                                 return -EINVAL;
 606                         if (copy_to_user(optval, address, len))
 607                                 return -EFAULT;
 608                         goto lenout;
 609                 }
 610
 611                 /* Dubious BSD thing... Probably nobody even uses it, but
 612                  * the UNIX standard wants it for whatever reason... -DaveM
 613                  */
 614                 case SO_ACCEPTCONN:
 615                         v.val = sk->sk_state == TCP_LISTEN;
 616                         break;
 617
 618                 case SO_PEERSEC:
 619                         return security_socket_getpeersec(sock, optval, optlen, len);
 620
 621                 default:
 622                         return(-ENOPROTOOPT);
 623         }
 624         if (len > lv)
 625                 len = lv;
 626         if (copy_to_user(optval, &v, len))
 627                 return -EFAULT;
 628 lenout:
 629         if (put_user(len, optlen))
 630                 return -EFAULT;
 631         return 0;
 632 }
 633
 634 /**
 635  *      sk_alloc - All socket objects are allocated here
 636  *      @family: protocol family
 637  *      @priority: for allocation (%GFP_KERNEL, %GFP_ATOMIC, etc)
 638  *      @prot: struct proto associated with this new sock instance
 639  *      @zero_it: if we should zero the newly allocated sock
 640  */
 641 struct sock *sk_alloc(int family, gfp_t priority,
 642                       struct proto *prot, int zero_it)
 643 {
 644         struct sock *sk = NULL;
 645         kmem_cache_t *slab = prot->slab;
 646
 647         if (slab != NULL)
 648                 sk = kmem_cache_alloc(slab, priority);
 649         else
 650                 sk = kmalloc(prot->obj_size, priority);
 651
 652         if (sk) {
 653                 if (zero_it) {
 654                         memset(sk, 0, prot->obj_size);
 655                         sk->sk_family = family;
 656                         /*
 657                          * See comment in struct sock definition to understand
 658                          * why we need sk_prot_creator -acme
 659                          */
 660                         sk->sk_prot = sk->sk_prot_creator = prot;
 661                         sock_lock_init(sk);
 662                 }
 663
 664                 if (security_sk_alloc(sk, family, priority))
 665                         goto out_free;
 666
 667                 if (!try_module_get(prot->owner))
 668                         goto out_free;
 669         }
 670         return sk;
 671
 672 out_free:
 673         if (slab != NULL)
 674                 kmem_cache_free(slab, sk);
 675         else
 676                 kfree(sk);
 677         return NULL;
 678 }
 679
 680 void sk_free(struct sock *sk)
 681 {
 682         struct sk_filter *filter;
 683         struct module *owner = sk->sk_prot_creator->owner;
 684
 685         if (sk->sk_destruct)
 686                 sk->sk_destruct(sk);
 687
 688         filter = sk->sk_filter;
 689         if (filter) {
 690                 sk_filter_release(sk, filter);
 691                 sk->sk_filter = NULL;
 692         }
 693
 694         sock_disable_timestamp(sk);
 695
 696         if (atomic_read(&sk->sk_omem_alloc))
 697                 printk(KERN_DEBUG "%s: optmem leakage (%d bytes) detected.\n",
 698                        __FUNCTION__, atomic_read(&sk->sk_omem_alloc));
 699
 700         security_sk_free(sk);
 701         if (sk->sk_prot_creator->slab != NULL)
 702                 kmem_cache_free(sk->sk_prot_creator->slab, sk);
 703         else
 704                 kfree(sk);
 705         module_put(owner);
 706 }
 707
 708 struct sock *sk_clone(const struct sock *sk, const gfp_t priority)
 709 {
 710         struct sock *newsk = sk_alloc(sk->sk_family, priority, sk->sk_prot, 0);
 711
 712         if (newsk != NULL) {
 713                 struct sk_filter *filter;
 714
 715                 memcpy(newsk, sk, sk->sk_prot->obj_size);
 716
 717                 /* SANITY */
 718                 sk_node_init(&newsk->sk_node);
 719                 sock_lock_init(newsk);
 720                 bh_lock_sock(newsk);
 721
 722                 atomic_set(&newsk->sk_rmem_alloc, 0);
 723                 atomic_set(&newsk->sk_wmem_alloc, 0);
 724                 atomic_set(&newsk->sk_omem_alloc, 0);
 725                 skb_queue_head_init(&newsk->sk_receive_queue);
 726                 skb_queue_head_init(&newsk->sk_write_queue);
 727
 728                 rwlock_init(&newsk->sk_dst_lock);
 729                 rwlock_init(&newsk->sk_callback_lock);
 730
 731                 newsk->sk_dst_cache     = NULL;
 732                 newsk->sk_wmem_queued   = 0;
 733                 newsk->sk_forward_alloc = 0;
 734                 newsk->sk_send_head     = NULL;
 735                 newsk->sk_backlog.head  = newsk->sk_backlog.tail = NULL;
 736                 newsk->sk_userlocks     = sk->sk_userlocks & ~SOCK_BINDPORT_LOCK;
 737
 738                 sock_reset_flag(newsk, SOCK_DONE);
 739                 skb_queue_head_init(&newsk->sk_error_queue);
 740
 741                 filter = newsk->sk_filter;
 742                 if (filter != NULL)
 743                         sk_filter_charge(newsk, filter);
 744
 745                 if (unlikely(xfrm_sk_clone_policy(newsk))) {
 746                         /* It is still raw copy of parent, so invalidate
 747                          * destructor and make plain sk_free() */
 748                         newsk->sk_destruct = NULL;
 749                         sk_free(newsk);
 750                         newsk = NULL;
 751                         goto out;
 752                 }
 753
 754                 newsk->sk_err      = 0;
 755                 newsk->sk_priority = 0;
 756                 atomic_set(&newsk->sk_refcnt, 2);
 757
 758                 /*
 759                  * Increment the counter in the same struct proto as the master
 760                  * sock (sk_refcnt_debug_inc uses newsk->sk_prot->socks, that
 761                  * is the same as sk->sk_prot->socks, as this field was copied
 762                  * with memcpy).
 763                  *
 764                  * This _changes_ the previous behaviour, where
 765                  * tcp_create_openreq_child always was incrementing the
 766                  * equivalent to tcp_prot->socks (inet_sock_nr), so this have
 767                  * to be taken into account in all callers. -acme
 768                  */
 769                 sk_refcnt_debug_inc(newsk);
 770                 newsk->sk_socket = NULL;
 771                 newsk->sk_sleep  = NULL;
 772
 773                 if (newsk->sk_prot->sockets_allocated)
 774                         atomic_inc(newsk->sk_prot->sockets_allocated);
 775         }
 776 out:
 777         return newsk;
 778 }
 779
 780 EXPORT_SYMBOL_GPL(sk_clone);
 781
 782 void __init sk_init(void)
 783 {
 784         if (num_physpages <= 4096) {
 785                 sysctl_wmem_max = 32767;
 786                 sysctl_rmem_max = 32767;
 787                 sysctl_wmem_default = 32767;
 788                 sysctl_rmem_default = 32767;
 789         } else if (num_physpages >= 131072) {
 790                 sysctl_wmem_max = 131071;
 791                 sysctl_rmem_max = 131071;
 792         }
 793 }
 794
 795 /*
 796  *      Simple resource managers for sockets.
 797  */
 798
 799
 800 /*
 801  * Write buffer destructor automatically called from kfree_skb.
 802  */
 803 void sock_wfree(struct sk_buff *skb)
 804 {
 805         struct sock *sk = skb->sk;
 806
 807         /* In case it might be waiting for more memory. */
 808         atomic_sub(skb->truesize, &sk->sk_wmem_alloc);
 809         if (!sock_flag(sk, SOCK_USE_WRITE_QUEUE))
 810                 sk->sk_write_space(sk);
 811         sock_put(sk);
 812 }
 813
 814 /*
 815  * Read buffer destructor automatically called from kfree_skb.
 816  */
 817 void sock_rfree(struct sk_buff *skb)
 818 {
 819         struct sock *sk = skb->sk;
 820
 821         atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
 822 }
 823
 824
 825 int sock_i_uid(struct sock *sk)
 826 {
 827         int uid;
 828
 829         read_lock(&sk->sk_callback_lock);
 830         uid = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_uid : 0;
 831         read_unlock(&sk->sk_callback_lock);
 832         return uid;
 833 }
 834
 835 unsigned long sock_i_ino(struct sock *sk)
 836 {
 837         unsigned long ino;
 838
 839         read_lock(&sk->sk_callback_lock);
 840         ino = sk->sk_socket ? SOCK_INODE(sk->sk_socket)->i_ino : 0;
 841         read_unlock(&sk->sk_callback_lock);
 842         return ino;
 843 }
 844
 845 /*
 846  * Allocate a skb from the socket's send buffer.
 847  */
 848 struct sk_buff *sock_wmalloc(struct sock *sk, unsigned long size, int force,
 849                              gfp_t priority)
 850 {
 851         if (force || atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 852                 struct sk_buff * skb = alloc_skb(size, priority);
 853                 if (skb) {
 854                         skb_set_owner_w(skb, sk);
 855                         return skb;
 856                 }
 857         }
 858         return NULL;
 859 }
 860
 861 /*
 862  * Allocate a skb from the socket's receive buffer.
 863  */
 864 struct sk_buff *sock_rmalloc(struct sock *sk, unsigned long size, int force,
 865                              gfp_t priority)
 866 {
 867         if (force || atomic_read(&sk->sk_rmem_alloc) < sk->sk_rcvbuf) {
 868                 struct sk_buff *skb = alloc_skb(size, priority);
 869                 if (skb) {
 870                         skb_set_owner_r(skb, sk);
 871                         return skb;
 872                 }
 873         }
 874         return NULL;
 875 }
 876
 877 /*
 878  * Allocate a memory block from the socket's option memory buffer.
 879  */
 880 void *sock_kmalloc(struct sock *sk, int size, gfp_t priority)
 881 {
 882         if ((unsigned)size <= sysctl_optmem_max &&
 883             atomic_read(&sk->sk_omem_alloc) + size < sysctl_optmem_max) {
 884                 void *mem;
 885                 /* First do the add, to avoid the race if kmalloc
 886                  * might sleep.
 887                  */
 888                 atomic_add(size, &sk->sk_omem_alloc);
 889                 mem = kmalloc(size, priority);
 890                 if (mem)
 891                         return mem;
 892                 atomic_sub(size, &sk->sk_omem_alloc);
 893         }
 894         return NULL;
 895 }
 896
 897 /*
 898  * Free an option memory block.
 899  */
 900 void sock_kfree_s(struct sock *sk, void *mem, int size)
 901 {
 902         kfree(mem);
 903         atomic_sub(size, &sk->sk_omem_alloc);
 904 }
 905
 906 /* It is almost wait_for_tcp_memory minus release_sock/lock_sock.
 907    I think, these locks should be removed for datagram sockets.
 908  */
 909 static long sock_wait_for_wmem(struct sock * sk, long timeo)
 910 {
 911         DEFINE_WAIT(wait);
 912
 913         clear_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 914         for (;;) {
 915                 if (!timeo)
 916                         break;
 917                 if (signal_pending(current))
 918                         break;
 919                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 920                 prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
 921                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf)
 922                         break;
 923                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 924                         break;
 925                 if (sk->sk_err)
 926                         break;
 927                 timeo = schedule_timeout(timeo);
 928         }
 929         finish_wait(sk->sk_sleep, &wait);
 930         return timeo;
 931 }
 932
 933
 934 /*
 935  *      Generic send/receive buffer handlers
 936  */
 937
 938 static struct sk_buff *sock_alloc_send_pskb(struct sock *sk,
 939                                             unsigned long header_len,
 940                                             unsigned long data_len,
 941                                             int noblock, int *errcode)
 942 {
 943         struct sk_buff *skb;
 944         gfp_t gfp_mask;
 945         long timeo;
 946         int err;
 947
 948         gfp_mask = sk->sk_allocation;
 949         if (gfp_mask & __GFP_WAIT)
 950                 gfp_mask |= __GFP_REPEAT;
 951
 952         timeo = sock_sndtimeo(sk, noblock);
 953         while (1) {
 954                 err = sock_error(sk);
 955                 if (err != 0)
 956                         goto failure;
 957
 958                 err = -EPIPE;
 959                 if (sk->sk_shutdown & SEND_SHUTDOWN)
 960                         goto failure;
 961
 962                 if (atomic_read(&sk->sk_wmem_alloc) < sk->sk_sndbuf) {
 963                         skb = alloc_skb(header_len, sk->sk_allocation);
 964                         if (skb) {
 965                                 int npages;
 966                                 int i;
 967
 968                                 /* No pages, we're done... */
 969                                 if (!data_len)
 970                                         break;
 971
 972                                 npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 973                                 skb->truesize += data_len;
 974                                 skb_shinfo(skb)->nr_frags = npages;
 975                                 for (i = 0; i < npages; i++) {
 976                                         struct page *page;
 977                                         skb_frag_t *frag;
 978
 979                                         page = alloc_pages(sk->sk_allocation, 0);
 980                                         if (!page) {
 981                                                 err = -ENOBUFS;
 982                                                 skb_shinfo(skb)->nr_frags = i;
 983                                                 kfree_skb(skb);
 984                                                 goto failure;
 985                                         }
 986
 987                                         frag = &skb_shinfo(skb)->frags[i];
 988                                         frag->page = page;
 989                                         frag->page_offset = 0;
 990                                         frag->size = (data_len >= PAGE_SIZE ?
 991                                                       PAGE_SIZE :
 992                                                       data_len);
 993                                         data_len -= PAGE_SIZE;
 994                                 }
 995
 996                                 /* Full success... */
 997                                 break;
 998                         }
 999                         err = -ENOBUFS;
1000                         goto failure;
1001                 }
1002                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
1003                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1004                 err = -EAGAIN;
1005                 if (!timeo)
1006                         goto failure;
1007                 if (signal_pending(current))
1008                         goto interrupted;
1009                 timeo = sock_wait_for_wmem(sk, timeo);
1010         }
1011
1012         skb_set_owner_w(skb, sk);
1013         return skb;
1014
1015 interrupted:
1016         err = sock_intr_errno(timeo);
1017 failure:
1018         *errcode = err;
1019         return NULL;
1020 }
1021
1022 struct sk_buff *sock_alloc_send_skb(struct sock *sk, unsigned long size,
1023                                     int noblock, int *errcode)
1024 {
1025         return sock_alloc_send_pskb(sk, size, 0, noblock, errcode);
1026 }
1027
1028 static void __lock_sock(struct sock *sk)
1029 {
1030         DEFINE_WAIT(wait);
1031
1032         for(;;) {
1033                 prepare_to_wait_exclusive(&sk->sk_lock.wq, &wait,
1034                                         TASK_UNINTERRUPTIBLE);
1035                 spin_unlock_bh(&sk->sk_lock.slock);
1036                 schedule();
1037                 spin_lock_bh(&sk->sk_lock.slock);
1038                 if(!sock_owned_by_user(sk))
1039                         break;
1040         }
1041         finish_wait(&sk->sk_lock.wq, &wait);
1042 }
1043
1044 static void __release_sock(struct sock *sk)
1045 {
1046         struct sk_buff *skb = sk->sk_backlog.head;
1047
1048         do {
1049                 sk->sk_backlog.head = sk->sk_backlog.tail = NULL;
1050                 bh_unlock_sock(sk);
1051
1052                 do {
1053                         struct sk_buff *next = skb->next;
1054
1055                         skb->next = NULL;
1056                         sk->sk_backlog_rcv(sk, skb);
1057
1058                         /*
1059                          * We are in process context here with softirqs
1060                          * disabled, use cond_resched_softirq() to preempt.
1061                          * This is safe to do because we've taken the backlog
1062                          * queue private:
1063                          */
1064                         cond_resched_softirq();
1065
1066                         skb = next;
1067                 } while (skb != NULL);
1068
1069                 bh_lock_sock(sk);
1070         } while((skb = sk->sk_backlog.head) != NULL);
1071 }
1072
1073 /**
1074  * sk_wait_data - wait for data to arrive at sk_receive_queue
1075  * @sk:    sock to wait on
1076  * @timeo: for how long
1077  *
1078  * Now socket state including sk->sk_err is changed only under lock,
1079  * hence we may omit checks after joining wait queue.
1080  * We check receive queue before schedule() only as optimization;
1081  * it is very likely that release_sock() added new data.
1082  */
1083 int sk_wait_data(struct sock *sk, long *timeo)
1084 {
1085         int rc;
1086         DEFINE_WAIT(wait);
1087
1088         prepare_to_wait(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
1089         set_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1090         rc = sk_wait_event(sk, timeo, !skb_queue_empty(&sk->sk_receive_queue));
1091         clear_bit(SOCK_ASYNC_WAITDATA, &sk->sk_socket->flags);
1092         finish_wait(sk->sk_sleep, &wait);
1093         return rc;
1094 }
1095
1096 EXPORT_SYMBOL(sk_wait_data);
1097
1098 /*
1099  * Set of default routines for initialising struct proto_ops when
1100  * the protocol does not support a particular function. In certain
1101  * cases where it makes no sense for a protocol to have a "do nothing"
1102  * function, some default processing is provided.
1103  */
1104
1105 int sock_no_bind(struct socket *sock, struct sockaddr *saddr, int len)
1106 {
1107         return -EOPNOTSUPP;
1108 }
1109
1110 int sock_no_connect(struct socket *sock, struct sockaddr *saddr,
1111                     int len, int flags)
1112 {
1113         return -EOPNOTSUPP;
1114 }
1115
1116 int sock_no_socketpair(struct socket *sock1, struct socket *sock2)
1117 {
1118         return -EOPNOTSUPP;
1119 }
1120
1121 int sock_no_accept(struct socket *sock, struct socket *newsock, int flags)
1122 {
1123         return -EOPNOTSUPP;
1124 }
1125
1126 int sock_no_getname(struct socket *sock, struct sockaddr *saddr,
1127                     int *len, int peer)
1128 {
1129         return -EOPNOTSUPP;
1130 }
1131
1132 unsigned int sock_no_poll(struct file * file, struct socket *sock, poll_table *pt)
1133 {
1134         return 0;
1135 }
1136
1137 int sock_no_ioctl(struct socket *sock, unsigned int cmd, unsigned long arg)
1138 {
1139         return -EOPNOTSUPP;
1140 }
1141
1142 int sock_no_listen(struct socket *sock, int backlog)
1143 {
1144         return -EOPNOTSUPP;
1145 }
1146
1147 int sock_no_shutdown(struct socket *sock, int how)
1148 {
1149         return -EOPNOTSUPP;
1150 }
1151
1152 int sock_no_setsockopt(struct socket *sock, int level, int optname,
1153                     char __user *optval, int optlen)
1154 {
1155         return -EOPNOTSUPP;
1156 }
1157
1158 int sock_no_getsockopt(struct socket *sock, int level, int optname,
1159                     char __user *optval, int __user *optlen)
1160 {
1161         return -EOPNOTSUPP;
1162 }
1163
1164 int sock_no_sendmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1165                     size_t len)
1166 {
1167         return -EOPNOTSUPP;
1168 }
1169
1170 int sock_no_recvmsg(struct kiocb *iocb, struct socket *sock, struct msghdr *m,
1171                     size_t len, int flags)
1172 {
1173         return -EOPNOTSUPP;
1174 }
1175
1176 int sock_no_mmap(struct file *file, struct socket *sock, struct vm_area_struct *vma)
1177 {
1178         /* Mirror missing mmap method error code */
1179         return -ENODEV;
1180 }
1181
1182 ssize_t sock_no_sendpage(struct socket *sock, struct page *page, int offset, size_t size, int flags)
1183 {
1184         ssize_t res;
1185         struct msghdr msg = {.msg_flags = flags};
1186         struct kvec iov;
1187         char *kaddr = kmap(page);
1188         iov.iov_base = kaddr + offset;
1189         iov.iov_len = size;
1190         res = kernel_sendmsg(sock, &msg, &iov, 1, size);
1191         kunmap(page);
1192         return res;
1193 }
1194
1195 /*
1196  *      Default Socket Callbacks
1197  */
1198
1199 static void sock_def_wakeup(struct sock *sk)
1200 {
1201         read_lock(&sk->sk_callback_lock);
1202         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1203                 wake_up_interruptible_all(sk->sk_sleep);
1204         read_unlock(&sk->sk_callback_lock);
1205 }
1206
1207 static void sock_def_error_report(struct sock *sk)
1208 {
1209         read_lock(&sk->sk_callback_lock);
1210         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1211                 wake_up_interruptible(sk->sk_sleep);
1212         sk_wake_async(sk,0,POLL_ERR);
1213         read_unlock(&sk->sk_callback_lock);
1214 }
1215
1216 static void sock_def_readable(struct sock *sk, int len)
1217 {
1218         read_lock(&sk->sk_callback_lock);
1219         if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1220                 wake_up_interruptible(sk->sk_sleep);
1221         sk_wake_async(sk,1,POLL_IN);
1222         read_unlock(&sk->sk_callback_lock);
1223 }
1224
1225 static void sock_def_write_space(struct sock *sk)
1226 {
1227         read_lock(&sk->sk_callback_lock);
1228
1229         /* Do not wake up a writer until he can make "significant"
1230          * progress.  --DaveM
1231          */
1232         if((atomic_read(&sk->sk_wmem_alloc) << 1) <= sk->sk_sndbuf) {
1233                 if (sk->sk_sleep && waitqueue_active(sk->sk_sleep))
1234                         wake_up_interruptible(sk->sk_sleep);
1235
1236                 /* Should agree with poll, otherwise some programs break */
1237                 if (sock_writeable(sk))
1238                         sk_wake_async(sk, 2, POLL_OUT);
1239         }
1240
1241         read_unlock(&sk->sk_callback_lock);
1242 }
1243
1244 static void sock_def_destruct(struct sock *sk)
1245 {
1246         kfree(sk->sk_protinfo);
1247 }
1248
1249 void sk_send_sigurg(struct sock *sk)
1250 {
1251         if (sk->sk_socket && sk->sk_socket->file)
1252                 if (send_sigurg(&sk->sk_socket->file->f_owner))
1253                         sk_wake_async(sk, 3, POLL_PRI);
1254 }
1255
1256 void sk_reset_timer(struct sock *sk, struct timer_list* timer,
1257                     unsigned long expires)
1258 {
1259         if (!mod_timer(timer, expires))
1260                 sock_hold(sk);
1261 }
1262
1263 EXPORT_SYMBOL(sk_reset_timer);
1264
1265 void sk_stop_timer(struct sock *sk, struct timer_list* timer)
1266 {
1267         if (timer_pending(timer) && del_timer(timer))
1268                 __sock_put(sk);
1269 }
1270
1271 EXPORT_SYMBOL(sk_stop_timer);
1272
1273 void sock_init_data(struct socket *sock, struct sock *sk)
1274 {
1275         skb_queue_head_init(&sk->sk_receive_queue);
1276         skb_queue_head_init(&sk->sk_write_queue);
1277         skb_queue_head_init(&sk->sk_error_queue);
1278
1279         sk->sk_send_head        =       NULL;
1280
1281         init_timer(&sk->sk_timer);
1282
1283         sk->sk_allocation       =       GFP_KERNEL;
1284         sk->sk_rcvbuf           =       sysctl_rmem_default;
1285         sk->sk_sndbuf           =       sysctl_wmem_default;
1286         sk->sk_state            =       TCP_CLOSE;
1287         sk->sk_socket           =       sock;
1288
1289         sock_set_flag(sk, SOCK_ZAPPED);
1290
1291         if(sock)
1292         {
1293                 sk->sk_type     =       sock->type;
1294                 sk->sk_sleep    =       &sock->wait;
1295                 sock->sk        =       sk;
1296         } else
1297                 sk->sk_sleep    =       NULL;
1298
1299         rwlock_init(&sk->sk_dst_lock);
1300         rwlock_init(&sk->sk_callback_lock);
1301
1302         sk->sk_state_change     =       sock_def_wakeup;
1303         sk->sk_data_ready       =       sock_def_readable;
1304         sk->sk_write_space      =       sock_def_write_space;
1305         sk->sk_error_report     =       sock_def_error_report;
1306         sk->sk_destruct         =       sock_def_destruct;
1307
1308         sk->sk_sndmsg_page      =       NULL;
1309         sk->sk_sndmsg_off       =       0;
1310
1311         sk->sk_peercred.pid     =       0;
1312         sk->sk_peercred.uid     =       -1;
1313         sk->sk_peercred.gid     =       -1;
1314         sk->sk_write_pending    =       0;
1315         sk->sk_rcvlowat         =       1;
1316         sk->sk_rcvtimeo         =       MAX_SCHEDULE_TIMEOUT;
1317         sk->sk_sndtimeo         =       MAX_SCHEDULE_TIMEOUT;
1318
1319         sk->sk_stamp.tv_sec     = -1L;
1320         sk->sk_stamp.tv_usec    = -1L;
1321
1322         atomic_set(&sk->sk_refcnt, 1);
1323 }
1324
1325 void fastcall lock_sock(struct sock *sk)
1326 {
1327         might_sleep();
1328         spin_lock_bh(&(sk->sk_lock.slock));
1329         if (sk->sk_lock.owner)
1330                 __lock_sock(sk);
1331         sk->sk_lock.owner = (void *)1;
1332         spin_unlock_bh(&(sk->sk_lock.slock));
1333 }
1334
1335 EXPORT_SYMBOL(lock_sock);
1336
1337 void fastcall release_sock(struct sock *sk)
1338 {
1339         spin_lock_bh(&(sk->sk_lock.slock));
1340         if (sk->sk_backlog.tail)
1341                 __release_sock(sk);
1342         sk->sk_lock.owner = NULL;
1343         if (waitqueue_active(&(sk->sk_lock.wq)))
1344                 wake_up(&(sk->sk_lock.wq));
1345         spin_unlock_bh(&(sk->sk_lock.slock));
1346 }
1347 EXPORT_SYMBOL(release_sock);
1348
1349 int sock_get_timestamp(struct sock *sk, struct timeval __user *userstamp)
1350 {
1351         if (!sock_flag(sk, SOCK_TIMESTAMP))
1352                 sock_enable_timestamp(sk);
1353         if (sk->sk_stamp.tv_sec == -1)
1354                 return -ENOENT;
1355         if (sk->sk_stamp.tv_sec == 0)
1356                 do_gettimeofday(&sk->sk_stamp);
1357         return copy_to_user(userstamp, &sk->sk_stamp, sizeof(struct timeval)) ?
1358                 -EFAULT : 0;
1359 }
1360 EXPORT_SYMBOL(sock_get_timestamp);
1361
1362 void sock_enable_timestamp(struct sock *sk)
1363 {
1364         if (!sock_flag(sk, SOCK_TIMESTAMP)) {
1365                 sock_set_flag(sk, SOCK_TIMESTAMP);
1366                 net_enable_timestamp();
1367         }
1368 }
1369 EXPORT_SYMBOL(sock_enable_timestamp);
1370
1371 /*
1372  *      Get a socket option on an socket.
1373  *
1374  *      FIX: POSIX 1003.1g is very ambiguous here. It states that
1375  *      asynchronous errors should be reported by getsockopt. We assume
1376  *      this means if you specify SO_ERROR (otherwise whats the point of it).
1377  */
1378 int sock_common_getsockopt(struct socket *sock, int level, int optname,
1379                            char __user *optval, int __user *optlen)
1380 {
1381         struct sock *sk = sock->sk;
1382
1383         return sk->sk_prot->getsockopt(sk, level, optname, optval, optlen);
1384 }
1385
1386 EXPORT_SYMBOL(sock_common_getsockopt);
1387
1388 int sock_common_recvmsg(struct kiocb *iocb, struct socket *sock,
1389                         struct msghdr *msg, size_t size, int flags)
1390 {
1391         struct sock *sk = sock->sk;
1392         int addr_len = 0;
1393         int err;
1394
1395         err = sk->sk_prot->recvmsg(iocb, sk, msg, size, flags & MSG_DONTWAIT,
1396                                    flags & ~MSG_DONTWAIT, &addr_len);
1397         if (err >= 0)
1398                 msg->msg_namelen = addr_len;
1399         return err;
1400 }
1401
1402 EXPORT_SYMBOL(sock_common_recvmsg);
1403
1404 /*
1405  *      Set socket options on an inet socket.
1406  */
1407 int sock_common_setsockopt(struct socket *sock, int level, int optname,
1408                            char __user *optval, int optlen)
1409 {
1410         struct sock *sk = sock->sk;
1411
1412         return sk->sk_prot->setsockopt(sk, level, optname, optval, optlen);
1413 }
1414
1415 EXPORT_SYMBOL(sock_common_setsockopt);
1416
1417 void sk_common_release(struct sock *sk)
1418 {
1419         if (sk->sk_prot->destroy)
1420                 sk->sk_prot->destroy(sk);
1421
1422         /*
1423          * Observation: when sock_common_release is called, processes have
1424          * no access to socket. But net still has.
1425          * Step one, detach it from networking:
1426          *
1427          * A. Remove from hash tables.
1428          */
1429
1430         sk->sk_prot->unhash(sk);
1431
1432         /*
1433          * In this point socket cannot receive new packets, but it is possible
1434          * that some packets are in flight because some CPU runs receiver and
1435          * did hash table lookup before we unhashed socket. They will achieve
1436          * receive queue and will be purged by socket destructor.
1437          *
1438          * Also we still have packets pending on receive queue and probably,
1439          * our own packets waiting in device queues. sock_destroy will drain
1440          * receive queue, but transmitted packets will delay socket destruction
1441          * until the last reference will be released.
1442          */
1443
1444         sock_orphan(sk);
1445
1446         xfrm_sk_free_policy(sk);
1447
1448         sk_refcnt_debug_release(sk);
1449         sock_put(sk);
1450 }
1451
1452 EXPORT_SYMBOL(sk_common_release);
1453
1454 static DEFINE_RWLOCK(proto_list_lock);
1455 static LIST_HEAD(proto_list);
1456
1457 int proto_register(struct proto *prot, int alloc_slab)
1458 {
1459         char *request_sock_slab_name = NULL;
1460         char *timewait_sock_slab_name;
1461         int rc = -ENOBUFS;
1462
1463         if (alloc_slab) {
1464                 prot->slab = kmem_cache_create(prot->name, prot->obj_size, 0,
1465                                                SLAB_HWCACHE_ALIGN, NULL, NULL);
1466
1467                 if (prot->slab == NULL) {
1468                         printk(KERN_CRIT "%s: Can't create sock SLAB cache!\n",
1469                                prot->name);
1470                         goto out;
1471                 }
1472
1473                 if (prot->rsk_prot != NULL) {
1474                         static const char mask[] = "request_sock_%s";
1475
1476                         request_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1477                         if (request_sock_slab_name == NULL)
1478                                 goto out_free_sock_slab;
1479
1480                         sprintf(request_sock_slab_name, mask, prot->name);
1481                         prot->rsk_prot->slab = kmem_cache_create(request_sock_slab_name,
1482                                                                  prot->rsk_prot->obj_size, 0,
1483                                                                  SLAB_HWCACHE_ALIGN, NULL, NULL);
1484
1485                         if (prot->rsk_prot->slab == NULL) {
1486                                 printk(KERN_CRIT "%s: Can't create request sock SLAB cache!\n",
1487                                        prot->name);
1488                                 goto out_free_request_sock_slab_name;
1489                         }
1490                 }
1491
1492                 if (prot->twsk_prot != NULL) {
1493                         static const char mask[] = "tw_sock_%s";
1494
1495                         timewait_sock_slab_name = kmalloc(strlen(prot->name) + sizeof(mask) - 1, GFP_KERNEL);
1496
1497                         if (timewait_sock_slab_name == NULL)
1498                                 goto out_free_request_sock_slab;
1499
1500                         sprintf(timewait_sock_slab_name, mask, prot->name);
1501                         prot->twsk_prot->twsk_slab =
1502                                 kmem_cache_create(timewait_sock_slab_name,
1503                                                   prot->twsk_prot->twsk_obj_size,
1504                                                   0, SLAB_HWCACHE_ALIGN,
1505                                                   NULL, NULL);
1506                         if (prot->twsk_prot->twsk_slab == NULL)
1507                                 goto out_free_timewait_sock_slab_name;
1508                 }
1509         }
1510
1511         write_lock(&proto_list_lock);
1512         list_add(&prot->node, &proto_list);
1513         write_unlock(&proto_list_lock);
1514         rc = 0;
1515 out:
1516         return rc;
1517 out_free_timewait_sock_slab_name:
1518         kfree(timewait_sock_slab_name);
1519 out_free_request_sock_slab:
1520         if (prot->rsk_prot && prot->rsk_prot->slab) {
1521                 kmem_cache_destroy(prot->rsk_prot->slab);
1522                 prot->rsk_prot->slab = NULL;
1523         }
1524 out_free_request_sock_slab_name:
1525         kfree(request_sock_slab_name);
1526 out_free_sock_slab:
1527         kmem_cache_destroy(prot->slab);
1528         prot->slab = NULL;
1529         goto out;
1530 }
1531
1532 EXPORT_SYMBOL(proto_register);
1533
1534 void proto_unregister(struct proto *prot)
1535 {
1536         write_lock(&proto_list_lock);
1537         list_del(&prot->node);
1538         write_unlock(&proto_list_lock);
1539
1540         if (prot->slab != NULL) {
1541                 kmem_cache_destroy(prot->slab);
1542                 prot->slab = NULL;
1543         }
1544
1545         if (prot->rsk_prot != NULL && prot->rsk_prot->slab != NULL) {
1546                 const char *name = kmem_cache_name(prot->rsk_prot->slab);
1547
1548                 kmem_cache_destroy(prot->rsk_prot->slab);
1549                 kfree(name);
1550                 prot->rsk_prot->slab = NULL;
1551         }
1552
1553         if (prot->twsk_prot != NULL && prot->twsk_prot->twsk_slab != NULL) {
1554                 const char *name = kmem_cache_name(prot->twsk_prot->twsk_slab);
1555
1556                 kmem_cache_destroy(prot->twsk_prot->twsk_slab);
1557                 kfree(name);
1558                 prot->twsk_prot->twsk_slab = NULL;
1559         }
1560 }
1561
1562 EXPORT_SYMBOL(proto_unregister);
1563
1564 #ifdef CONFIG_PROC_FS
1565 static inline struct proto *__proto_head(void)
1566 {
1567         return list_entry(proto_list.next, struct proto, node);
1568 }
1569
1570 static inline struct proto *proto_head(void)
1571 {
1572         return list_empty(&proto_list) ? NULL : __proto_head();
1573 }
1574
1575 static inline struct proto *proto_next(struct proto *proto)
1576 {
1577         return proto->node.next == &proto_list ? NULL :
1578                 list_entry(proto->node.next, struct proto, node);
1579 }
1580
1581 static inline struct proto *proto_get_idx(loff_t pos)
1582 {
1583         struct proto *proto;
1584         loff_t i = 0;
1585
1586         list_for_each_entry(proto, &proto_list, node)
1587                 if (i++ == pos)
1588                         goto out;
1589
1590         proto = NULL;
1591 out:
1592         return proto;
1593 }
1594
1595 static void *proto_seq_start(struct seq_file *seq, loff_t *pos)
1596 {
1597         read_lock(&proto_list_lock);
1598         return *pos ? proto_get_idx(*pos - 1) : SEQ_START_TOKEN;
1599 }
1600
1601 static void *proto_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1602 {
1603         ++*pos;
1604         return v == SEQ_START_TOKEN ? proto_head() : proto_next(v);
1605 }
1606
1607 static void proto_seq_stop(struct seq_file *seq, void *v)
1608 {
1609         read_unlock(&proto_list_lock);
1610 }
1611
1612 static char proto_method_implemented(const void *method)
1613 {
1614         return method == NULL ? 'n' : 'y';
1615 }
1616
1617 static void proto_seq_printf(struct seq_file *seq, struct proto *proto)
1618 {
1619         seq_printf(seq, "%-9s %4u %6d  %6d   %-3s %6u   %-3s  %-10s "
1620                         "%2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c %2c\n",
1621                    proto->name,
1622                    proto->obj_size,
1623                    proto->sockets_allocated != NULL ? atomic_read(proto->sockets_allocated) : -1,
1624                    proto->memory_allocated != NULL ? atomic_read(proto->memory_allocated) : -1,
1625                    proto->memory_pressure != NULL ? *proto->memory_pressure ? "yes" : "no" : "NI",
1626                    proto->max_header,
1627                    proto->slab == NULL ? "no" : "yes",
1628                    module_name(proto->owner),
1629                    proto_method_implemented(proto->close),
1630                    proto_method_implemented(proto->connect),
1631                    proto_method_implemented(proto->disconnect),
1632                    proto_method_implemented(proto->accept),
1633                    proto_method_implemented(proto->ioctl),
1634                    proto_method_implemented(proto->init),
1635                    proto_method_implemented(proto->destroy),
1636                    proto_method_implemented(proto->shutdown),
1637                    proto_method_implemented(proto->setsockopt),
1638                    proto_method_implemented(proto->getsockopt),
1639                    proto_method_implemented(proto->sendmsg),
1640                    proto_method_implemented(proto->recvmsg),
1641                    proto_method_implemented(proto->sendpage),
1642                    proto_method_implemented(proto->bind),
1643                    proto_method_implemented(proto->backlog_rcv),
1644                    proto_method_implemented(proto->hash),
1645                    proto_method_implemented(proto->unhash),
1646                    proto_method_implemented(proto->get_port),
1647                    proto_method_implemented(proto->enter_memory_pressure));
1648 }
1649
1650 static int proto_seq_show(struct seq_file *seq, void *v)
1651 {
1652         if (v == SEQ_START_TOKEN)
1653                 seq_printf(seq, "%-9s %-4s %-8s %-6s %-5s %-7s %-4s %-10s %s",
1654                            "protocol",
1655                            "size",
1656                            "sockets",
1657                            "memory",
1658                            "press",
1659                            "maxhdr",
1660                            "slab",
1661                            "module",
1662                            "cl co di ac io in de sh ss gs se re sp bi br ha uh gp em\n");
1663         else
1664                 proto_seq_printf(seq, v);
1665         return 0;
1666 }
1667
1668 static struct seq_operations proto_seq_ops = {
1669         .start  = proto_seq_start,
1670         .next   = proto_seq_next,
1671         .stop   = proto_seq_stop,
1672         .show   = proto_seq_show,
1673 };
1674
1675 static int proto_seq_open(struct inode *inode, struct file *file)
1676 {
1677         return seq_open(file, &proto_seq_ops);
1678 }
1679
1680 static struct file_operations proto_seq_fops = {
1681         .owner          = THIS_MODULE,
1682         .open           = proto_seq_open,
1683         .read           = seq_read,
1684         .llseek         = seq_lseek,
1685         .release        = seq_release,
1686 };
1687
1688 static int __init proto_init(void)
1689 {
1690         /* register /proc/net/protocols */
1691         return proc_net_fops_create("protocols", S_IRUGO, &proto_seq_fops) == NULL ? -ENOBUFS : 0;
1692 }
1693
1694 subsys_initcall(proto_init);
1695
1696 #endif /* PROC_FS */
1697
1698 EXPORT_SYMBOL(sk_alloc);
1699 EXPORT_SYMBOL(sk_free);
1700 EXPORT_SYMBOL(sk_send_sigurg);
1701 EXPORT_SYMBOL(sock_alloc_send_skb);
1702 EXPORT_SYMBOL(sock_init_data);
1703 EXPORT_SYMBOL(sock_kfree_s);
1704 EXPORT_SYMBOL(sock_kmalloc);
1705 EXPORT_SYMBOL(sock_no_accept);
1706 EXPORT_SYMBOL(sock_no_bind);
1707 EXPORT_SYMBOL(sock_no_connect);
1708 EXPORT_SYMBOL(sock_no_getname);
1709 EXPORT_SYMBOL(sock_no_getsockopt);
1710 EXPORT_SYMBOL(sock_no_ioctl);
1711 EXPORT_SYMBOL(sock_no_listen);
1712 EXPORT_SYMBOL(sock_no_mmap);
1713 EXPORT_SYMBOL(sock_no_poll);
1714 EXPORT_SYMBOL(sock_no_recvmsg);
1715 EXPORT_SYMBOL(sock_no_sendmsg);
1716 EXPORT_SYMBOL(sock_no_sendpage);
1717 EXPORT_SYMBOL(sock_no_setsockopt);
1718 EXPORT_SYMBOL(sock_no_shutdown);
1719 EXPORT_SYMBOL(sock_no_socketpair);
1720 EXPORT_SYMBOL(sock_rfree);
1721 EXPORT_SYMBOL(sock_setsockopt);
1722 EXPORT_SYMBOL(sock_wfree);
1723 EXPORT_SYMBOL(sock_wmalloc);
1724 EXPORT_SYMBOL(sock_i_uid);
1725 EXPORT_SYMBOL(sock_i_ino);
1726 EXPORT_SYMBOL(sysctl_optmem_max);
1727 #ifdef CONFIG_SYSCTL
1728 EXPORT_SYMBOL(sysctl_rmem_max);
1729 EXPORT_SYMBOL(sysctl_wmem_max);
1730 #endif